Man, this looks ugly. I'm just waiting for Bill Irwin, or Anton to
trump me, though.
PID TTY STAT TIME COMMAND
1 ? S 0:10 init
2 ? SW 0:00 [migration/0]
3 ? SWN 0:00 [ksoftirqd/0]
4 ? SW 0:00 [migration/1]
5 ? SWN 0:00 [ksoftirqd/1]
6 ? SW 0:00 [migration/2]
7 ? RWN 0:00 [ksoftirqd/2]
8 ? SW 0:00 [migration/3]
9 ? SWN 0:00 [ksoftirqd/3]
10 ? SW 0:00 [migration/4]
11 ? SWN 0:00 [ksoftirqd/4]
12 ? SW 0:00 [migration/5]
13 ? SWN 0:00 [ksoftirqd/5]
14 ? SW 0:00 [migration/6]
15 ? RWN 0:00 [ksoftirqd/6]
16 ? SW 0:00 [migration/7]
17 ? SWN 0:00 [ksoftirqd/7]
18 ? SW 0:00 [events/0]
19 ? SW 0:00 [events/1]
20 ? SW 0:00 [events/2]
21 ? SW 0:00 [events/3]
22 ? SW 0:00 [events/4]
23 ? SW 0:00 [events/5]
24 ? SW 0:00 [events/6]
25 ? SW 0:00 [events/7]
26 ? SW 0:00 [kswapd0]
28 ? SW 0:00 [pdflush]
27 ? SW 0:00 [pdflush]
29 ? SW 0:00 [aio/0]
30 ? SW 0:00 [aio/1]
31 ? SW 0:00 [aio/2]
32 ? SW 0:00 [aio/3]
33 ? SW 0:00 [aio/4]
34 ? SW 0:00 [aio/5]
35 ? SW 0:00 [aio/6]
36 ? SW 0:00 [aio/7]
37 ? SW 0:00 [scsi_eh_0]
38 ? SW 0:00 [scsi_eh_1]
39 ? SW 0:00 [scsi_eh_2]
40 ? SW 0:00 [kseriod]
41 ? DW 0:00 [kjournald]
115 ? SW 0:00 [kjournald]
116 ? DW 0:00 [kjournald]
117 ? SW 0:00 [kjournald]
--
Dave Hansen
[email protected]
On Fri, Oct 11, 2002 at 08:36:22PM -0700, Dave Hansen wrote:
> Man, this looks ugly. I'm just waiting for Bill Irwin, or Anton to
> trump me, though.
UID PID PPID C STIME TTY TIME CMD
root 1 0 0 Oct09 ? 00:00:08 init
root 2 1 0 Oct09 ? 00:00:00 [migration_CPU0]
root 3 1 0 Oct09 ? 00:00:00 [ksoftirqd_CPU0]
root 4 1 0 Oct09 ? 00:00:00 [migration_CPU1]
root 5 1 0 Oct09 ? 00:00:00 [ksoftirqd_CPU1]
root 6 1 0 Oct09 ? 00:00:00 [migration_CPU2]
root 7 1 0 Oct09 ? 00:00:00 [ksoftirqd_CPU2]
root 8 1 0 Oct09 ? 00:00:00 [migration_CPU3]
root 9 1 0 Oct09 ? 00:00:00 [ksoftirqd_CPU3]
root 10 1 0 Oct09 ? 00:00:00 [migration_CPU4]
root 11 1 0 Oct09 ? 00:00:00 [ksoftirqd_CPU4]
root 12 1 0 Oct09 ? 00:00:00 [migration_CPU5]
root 13 1 0 Oct09 ? 00:00:00 [ksoftirqd_CPU5]
root 14 1 0 Oct09 ? 00:00:00 [migration_CPU6]
root 15 1 0 Oct09 ? 00:00:00 [ksoftirqd_CPU6]
root 16 1 0 Oct09 ? 00:00:00 [migration_CPU7]
root 17 1 0 Oct09 ? 00:00:00 [ksoftirqd_CPU7]
root 18 1 0 Oct09 ? 00:00:00 [migration_CPU8]
root 19 1 0 Oct09 ? 00:00:00 [ksoftirqd_CPU8]
root 20 1 0 Oct09 ? 00:00:00 [migration_CPU9]
root 21 1 0 Oct09 ? 00:00:00 [ksoftirqd_CPU9]
root 22 1 0 Oct09 ? 00:00:00 [migration_CPU10]
root 23 1 0 Oct09 ? 00:00:00 [ksoftirqd_CPU10]
root 24 1 0 Oct09 ? 00:00:00 [migration_CPU11]
root 25 1 0 Oct09 ? 00:00:00 [ksoftirqd_CPU11]
root 26 1 0 Oct09 ? 00:00:00 [migration_CPU12]
root 27 1 0 Oct09 ? 00:00:00 [ksoftirqd_CPU12]
root 28 1 0 Oct09 ? 00:00:00 [migration_CPU13]
root 29 1 0 Oct09 ? 00:00:00 [ksoftirqd_CPU13]
root 30 1 0 Oct09 ? 00:00:00 [migration_CPU14]
root 31 1 0 Oct09 ? 00:00:00 [ksoftirqd_CPU14]
root 32 1 0 Oct09 ? 00:00:00 [migration_CPU15]
root 33 1 0 Oct09 ? 00:00:00 [ksoftirqd_CPU15]
root 34 1 0 Oct09 ? 00:00:00 [events/0]
root 35 1 0 Oct09 ? 00:00:00 [events/1]
root 36 1 0 Oct09 ? 00:00:00 [events/2]
root 37 1 0 Oct09 ? 00:00:00 [events/3]
root 38 1 0 Oct09 ? 00:00:00 [events/4]
root 39 1 0 Oct09 ? 00:00:00 [events/5]
root 40 1 0 Oct09 ? 00:00:00 [events/6]
root 41 1 0 Oct09 ? 00:00:00 [events/7]
root 42 1 0 Oct09 ? 00:00:00 [events/8]
root 43 1 0 Oct09 ? 00:00:00 [events/9]
root 44 1 0 Oct09 ? 00:00:00 [events/10]
root 45 1 0 Oct09 ? 00:00:00 [events/11]
root 46 1 0 Oct09 ? 00:00:00 [events/12]
root 47 1 0 Oct09 ? 00:00:00 [events/13]
root 48 1 0 Oct09 ? 00:00:00 [events/14]
root 49 1 0 Oct09 ? 00:00:00 [events/15]
root 71 1 0 Oct09 ? 00:00:00 [pdflush]
root 70 1 0 Oct09 ? 00:02:24 [pdflush]
root 69 1 0 Oct09 ? 00:00:00 [kswapd0]
root 68 1 0 Oct09 ? 00:00:00 [kswapd1]
root 67 1 0 Oct09 ? 00:00:00 [kswapd2]
root 66 1 0 Oct09 ? 00:00:00 [kswapd3]
root 72 1 0 Oct09 ? 00:00:00 [aio/0]
root 73 1 0 Oct09 ? 00:00:00 [aio/1]
root 74 1 0 Oct09 ? 00:00:00 [aio/2]
root 75 1 0 Oct09 ? 00:00:00 [aio/3]
root 76 1 0 Oct09 ? 00:00:00 [aio/4]
root 77 1 0 Oct09 ? 00:00:00 [aio/5]
root 78 1 0 Oct09 ? 00:00:00 [aio/6]
root 79 1 0 Oct09 ? 00:00:00 [aio/7]
root 80 1 0 Oct09 ? 00:00:00 [aio/8]
root 81 1 0 Oct09 ? 00:00:00 [aio/9]
root 82 1 0 Oct09 ? 00:00:00 [aio/10]
root 83 1 0 Oct09 ? 00:00:00 [aio/11]
root 84 1 0 Oct09 ? 00:00:00 [aio/12]
root 85 1 0 Oct09 ? 00:00:00 [aio/13]
root 86 1 0 Oct09 ? 00:00:00 [aio/14]
root 87 1 0 Oct09 ? 00:00:00 [aio/15]
root 88 1 0 Oct09 ? 00:00:00 [scsi_eh_0]
root 264 1 0 Oct09 ? 00:00:00 /sbin/syslogd
root 267 1 0 Oct09 ? 00:00:01 /sbin/klogd
root 274 1 0 Oct09 ? 00:00:00 /usr/sbin/inetd
root 279 1 0 Oct09 ? 00:00:00 /usr/sbin/netserver
root 285 1 0 Oct09 ? 00:00:00 /usr/sbin/sshd
root 288 1 0 Oct09 ? 00:00:07 /usr/sbin/ntpd
root 291 1 0 Oct09 tty1 00:00:00 /sbin/getty 38400 tty1
root 292 1 0 Oct09 tty2 00:00:00 /sbin/getty 38400 tty2
root 293 1 0 Oct09 tty3 00:00:00 /sbin/getty 38400 tty3
root 294 1 0 Oct09 tty4 00:00:00 /sbin/getty 38400 tty4
root 295 1 0 Oct09 tty5 00:00:00 /sbin/getty 38400 tty5
root 296 1 0 Oct09 tty6 00:00:00 /sbin/getty 38400 tty6
root 297 1 0 Oct09 ttyS0 00:00:00 -bash
root 303 1 0 Oct09 ? 00:00:00 sshd -p 1976
root 9582 285 0 Oct10 ? 00:00:00 /usr/sbin/sshd
wli 9584 9582 0 Oct10 ? 00:00:01 /usr/sbin/sshd
wli 9585 9584 0 Oct10 pts/0 00:00:00 -zsh
root 9589 285 0 Oct10 ? 00:00:00 /usr/sbin/sshd
wli 9591 9589 0 Oct10 ? 00:00:00 /usr/sbin/sshd
wli 9592 9591 0 Oct10 pts/1 00:00:00 -zsh
root 5901 285 0 20:44 ? 00:00:00 /usr/sbin/sshd
wli 5903 5901 0 20:44 ? 00:00:00 /usr/sbin/sshd
wli 5904 5903 1 20:44 pts/2 00:00:00 -zsh
wli 5907 5904 0 20:44 pts/2 00:00:00 ps -fade
wli 5908 5904 0 20:44 pts/2 00:00:00 less
From: Dave Hansen <[email protected]>
Date: Fri, 11 Oct 2002 20:36:22 -0700
Man, this looks ugly. I'm just waiting for Bill Irwin, or Anton to
trump me, though.
We could make them threads of process 0 :-)
> Man, this looks ugly. I'm just waiting for Bill Irwin, or Anton to
> trump me, though.
Id like to oblige but I hear the list has a 100kB limit :)
Anton
On Fri, 11 Oct 2002, David S. Miller wrote:
[ 8 gazillion kernel threads ]
> We could make them threads of process 0 :-)
That was my first thought too, but on second thought I think we've
got an excessive amount of kernel threads and should do something
about that...
regards,
Rik
--
Bravely reimplemented by the knights who say "NIH".
http://www.surriel.com/ http://distro.conectiva.com/
Current spamtrap: <a href=mailto:"[email protected]">[email protected]</a>
On Sat, Oct 12, 2002 at 01:51:41PM +1000, Anton Blanchard wrote:
> Id like to oblige but I hear the list has a 100kB limit :)
> Anton
Bah! I'm at a competitive disadvantage because I've got a lesser
BITS_PER_LONG. No matter, NR_CPUS > BITS_PER_LONG shall be conquered
and the explosion of kernel threads will be quite visible (though
unfortunately probably post-freeze).
Cheers,
Bill
> Bah! I'm at a competitive disadvantage because I've got a lesser
> BITS_PER_LONG. No matter, NR_CPUS > BITS_PER_LONG shall be conquered
> and the explosion of kernel threads will be quite visible (though
> unfortunately probably post-freeze).
Speaking of which, the recent CONFIG_NR_CPUS addition shows just how
bloated all our [NR_CPU] structures are. We need to get serious about
using the per cpu data stuff. Going from 32 to 64 was over 500kB on my
ppc64 build.
Anton
From: Anton Blanchard <[email protected]>
Date: Sat, 12 Oct 2002 14:09:59 +1000
Speaking of which, the recent CONFIG_NR_CPUS addition shows just how
bloated all our [NR_CPU] structures are. We need to get serious about
using the per cpu data stuff. Going from 32 to 64 was over 500kB on my
ppc64 build.
In fact, thinking about this some more, we should make the ".per_cpu"
bits emit a table entry instead of some dummy object which takes up
space. The table entry would be in the special .per_cpu
section still but be just a size value.
We should do this on both SMP and non-SMP and it will shrink the
kernel image size in both cases.
I don't have the time to implement this so I'll shut up now :-)
On Sat, 2002-10-12 at 04:55, Rik van Riel wrote:
> On Fri, 11 Oct 2002, David S. Miller wrote:
>
> [ 8 gazillion kernel threads ]
>
> > We could make them threads of process 0 :-)
>
> That was my first thought too, but on second thought I think we've
> got an excessive amount of kernel threads and should do something
> about that...
Migration workqueues ?
On Fri, Oct 11, 2002 at 11:53:29PM -0700, David S. Miller wrote:
> In fact, thinking about this some more, we should make the ".per_cpu"
> bits emit a table entry instead of some dummy object which takes up
> space. The table entry would be in the special .per_cpu
> section still but be just a size value.
That's more complicated. Using the linker to help out with
layout is definitely helpful. If you want to omit the per-cpu
area from the kernel image, then arrange for it to be .bss.
r~
From: Richard Henderson <[email protected]>
Date: Sat, 12 Oct 2002 13:15:01 -0700
On Fri, Oct 11, 2002 at 11:53:29PM -0700, David S. Miller wrote:
> In fact, thinking about this some more, we should make the ".per_cpu"
> bits emit a table entry instead of some dummy object which takes up
> space. The table entry would be in the special .per_cpu
> section still but be just a size value.
That's more complicated.
Hmm, we put arbitrary tables of information into seperate elf sections
already. Consider the exception fixup mechanism for example. That's
the kind of thing I was thinking about.
Oh I see, you're saying that then getting at the things symbolically
will be painful. Yes it needs more thought.
If you want to omit the per-cpu area from the kernel image, then
arrange for it to be .bss.
Good idea, well on SMP it can be marked throw-away (ie. __init_data).
> Good idea, well on SMP it can be marked throw-away (ie. __init_data).
We could also only create per cpu data areas when cpu_possible() is
true, instead of NR_CPUS worth. That might be a little dangerous however.
Anton
On Sat, Oct 12, 2002 at 11:27:44PM -0700, David S. Miller wrote:
> Good idea, well on SMP it can be marked throw-away (ie. __init_data).
Or you could use it for cpu0.
r~
Anton Blanchard wrote:
>
> > Bah! I'm at a competitive disadvantage because I've got a lesser
> > BITS_PER_LONG. No matter, NR_CPUS > BITS_PER_LONG shall be conquered
> > and the explosion of kernel threads will be quite visible (though
> > unfortunately probably post-freeze).
>
> Speaking of which, the recent CONFIG_NR_CPUS addition shows just how
> bloated all our [NR_CPU] structures are. We need to get serious about
> using the per cpu data stuff. Going from 32 to 64 was over 500kB on my
> ppc64 build.
>
Half of which is in timer.c.
mnm:/usr/src/25> size kernel/timer.o
text data bss dec hex filename
4960 100 167648 172708 2a2a4 kernel/timer.o
That's with NR_CPUS=32. Show me yours.
Using the percpu stuff will not significantly reduce this. Some
new data structure might be needed.
> Half of which is in timer.c.
>
> mnm:/usr/src/25> size kernel/timer.o
> text data bss dec hex filename
> 4960 100 167648 172708 2a2a4 kernel/timer.o
>
> That's with NR_CPUS=32. Show me yours.
Its not like me to miss show and tell.
CONFIG_NR_CPUS=32:
text data bss dec hex filename
8488 267544 71392 347424 54d20 kernel/timer.o
CONFIG_NR_CPUS=64:
text data bss dec hex filename
8488 533784 137568 679840 a5fa0 kernel/timer.o
Ouch.
Anton
On Fri, Oct 11, 2002 at 08:36:22PM -0700, Dave Hansen wrote:
> Man, this looks ugly. I'm just waiting for Bill Irwin, or Anton to
> trump me, though.
If I may rehash, since it is O(cpus), hence 1MB of ZONE_NORMAL stacks...
UID PID PPID C STIME TTY TIME CMD
root 1 0 4 18:56 ? 00:00:09 init
root 2 1 0 18:56 ? 00:00:00 [migration/0]
root 3 1 0 18:56 ? 00:00:00 [ksoftirqd/0]
root 4 1 0 18:56 ? 00:00:00 [migration/1]
root 5 1 0 18:56 ? 00:00:00 [ksoftirqd/1]
root 6 1 0 18:56 ? 00:00:00 [migration/2]
root 7 1 0 18:56 ? 00:00:00 [ksoftirqd/2]
root 8 1 0 18:56 ? 00:00:00 [migration/3]
root 9 1 0 18:56 ? 00:00:00 [ksoftirqd/3]
root 10 1 0 18:56 ? 00:00:00 [migration/4]
root 11 1 0 18:56 ? 00:00:00 [ksoftirqd/4]
root 12 1 0 18:56 ? 00:00:00 [migration/5]
root 13 1 0 18:56 ? 00:00:00 [ksoftirqd/5]
root 14 1 0 18:56 ? 00:00:00 [migration/6]
root 15 1 0 18:56 ? 00:00:00 [ksoftirqd/6]
root 16 1 0 18:56 ? 00:00:00 [migration/7]
root 17 1 0 18:56 ? 00:00:00 [ksoftirqd/7]
root 18 1 0 18:56 ? 00:00:00 [migration/8]
root 19 1 0 18:56 ? 00:00:00 [ksoftirqd/8]
root 20 1 0 18:56 ? 00:00:00 [migration/9]
root 21 1 0 18:56 ? 00:00:00 [ksoftirqd/9]
root 22 1 0 18:56 ? 00:00:00 [migration/10]
root 23 1 0 18:56 ? 00:00:00 [ksoftirqd/10]
root 24 1 0 18:56 ? 00:00:00 [migration/11]
root 25 1 0 18:56 ? 00:00:00 [ksoftirqd/11]
root 26 1 0 18:56 ? 00:00:00 [migration/12]
root 27 1 0 18:56 ? 00:00:00 [ksoftirqd/12]
root 28 1 0 18:56 ? 00:00:00 [migration/13]
root 29 1 0 18:56 ? 00:00:00 [ksoftirqd/13]
root 30 1 0 18:56 ? 00:00:00 [migration/14]
root 31 1 0 18:56 ? 00:00:00 [ksoftirqd/14]
root 32 1 0 18:56 ? 00:00:00 [migration/15]
root 33 1 0 18:56 ? 00:00:00 [ksoftirqd/15]
root 34 1 0 18:56 ? 00:00:00 [migration/16]
root 35 1 0 18:56 ? 00:00:00 [ksoftirqd/16]
root 36 1 0 18:56 ? 00:00:00 [migration/17]
root 37 1 0 18:56 ? 00:00:00 [ksoftirqd/17]
root 38 1 0 18:56 ? 00:00:00 [migration/18]
root 39 1 0 18:56 ? 00:00:00 [ksoftirqd/18]
root 40 1 0 18:56 ? 00:00:00 [migration/19]
root 41 1 0 18:56 ? 00:00:00 [ksoftirqd/19]
root 42 1 0 18:56 ? 00:00:00 [migration/20]
root 43 1 0 18:56 ? 00:00:00 [ksoftirqd/20]
root 44 1 0 18:56 ? 00:00:00 [migration/21]
root 45 1 0 18:56 ? 00:00:00 [ksoftirqd/21]
root 46 1 0 18:56 ? 00:00:00 [migration/22]
root 47 1 0 18:56 ? 00:00:00 [ksoftirqd/22]
root 48 1 0 18:56 ? 00:00:00 [migration/23]
root 49 1 0 18:56 ? 00:00:00 [ksoftirqd/23]
root 50 1 0 18:56 ? 00:00:00 [migration/24]
root 51 1 0 18:56 ? 00:00:00 [ksoftirqd/24]
root 52 1 0 18:56 ? 00:00:00 [migration/25]
root 53 1 0 18:56 ? 00:00:00 [ksoftirqd/25]
root 54 1 0 18:56 ? 00:00:00 [migration/26]
root 55 1 0 18:56 ? 00:00:00 [ksoftirqd/26]
root 56 1 0 18:56 ? 00:00:00 [migration/27]
root 57 1 0 18:56 ? 00:00:00 [ksoftirqd/27]
root 58 1 0 18:56 ? 00:00:00 [migration/28]
root 59 1 0 18:56 ? 00:00:00 [ksoftirqd/28]
root 60 1 0 18:56 ? 00:00:00 [migration/29]
root 61 1 0 18:56 ? 00:00:00 [ksoftirqd/29]
root 62 1 0 18:56 ? 00:00:00 [migration/30]
root 63 1 0 18:56 ? 00:00:00 [ksoftirqd/30]
root 64 1 0 18:56 ? 00:00:00 [migration/31]
root 65 1 0 18:56 ? 00:00:00 [ksoftirqd/31]
root 66 1 0 18:56 ? 00:00:00 [events/0]
root 67 1 0 18:56 ? 00:00:00 [events/1]
root 68 1 0 18:56 ? 00:00:00 [events/2]
root 69 1 0 18:56 ? 00:00:00 [events/3]
root 70 1 0 18:56 ? 00:00:00 [events/4]
root 71 1 0 18:56 ? 00:00:00 [events/5]
root 72 1 0 18:56 ? 00:00:00 [events/6]
root 73 1 0 18:56 ? 00:00:00 [events/7]
root 74 1 0 18:56 ? 00:00:00 [events/8]
root 75 1 0 18:56 ? 00:00:00 [events/9]
root 76 1 0 18:56 ? 00:00:00 [events/10]
root 77 1 0 18:56 ? 00:00:00 [events/11]
root 78 1 0 18:56 ? 00:00:00 [events/12]
root 79 1 0 18:56 ? 00:00:00 [events/13]
root 80 1 0 18:56 ? 00:00:00 [events/14]
root 81 1 0 18:56 ? 00:00:00 [events/15]
root 82 1 0 18:56 ? 00:00:00 [events/16]
root 83 1 0 18:56 ? 00:00:00 [events/17]
root 84 1 0 18:56 ? 00:00:00 [events/18]
root 85 1 0 18:56 ? 00:00:00 [events/19]
root 86 1 0 18:56 ? 00:00:00 [events/20]
root 87 1 0 18:56 ? 00:00:00 [events/21]
root 88 1 0 18:56 ? 00:00:00 [events/22]
root 89 1 0 18:56 ? 00:00:00 [events/23]
root 90 1 0 18:56 ? 00:00:00 [events/24]
root 91 1 0 18:56 ? 00:00:00 [events/25]
root 92 1 0 18:56 ? 00:00:00 [events/26]
root 93 1 0 18:56 ? 00:00:00 [events/27]
root 94 1 0 18:56 ? 00:00:00 [events/28]
root 95 1 0 18:56 ? 00:00:00 [events/29]
root 96 1 0 18:56 ? 00:00:00 [events/30]
root 97 1 0 18:56 ? 00:00:00 [events/31]
root 105 1 0 18:56 ? 00:00:00 [kswapd0]
root 104 1 0 18:56 ? 00:00:00 [kswapd1]
root 103 1 0 18:56 ? 00:00:00 [kswapd2]
root 102 1 0 18:56 ? 00:00:00 [kswapd3]
root 100 1 0 18:56 ? 00:00:00 [kswapd5]
root 101 1 0 18:56 ? 00:00:00 [kswapd4]
root 99 1 0 18:56 ? 00:00:00 [kswapd6]
root 98 1 0 18:56 ? 00:00:00 [kswapd7]
root 106 1 1 18:56 ? 00:00:03 [pdflush]
root 107 1 1 18:56 ? 00:00:03 [pdflush]
root 108 1 0 18:56 ? 00:00:00 [aio/0]
root 109 1 0 18:56 ? 00:00:00 [aio/1]
root 110 1 0 18:56 ? 00:00:00 [aio/2]
root 111 1 0 18:56 ? 00:00:00 [aio/3]
root 112 1 0 18:56 ? 00:00:00 [aio/4]
root 113 1 0 18:56 ? 00:00:00 [aio/5]
root 114 1 0 18:56 ? 00:00:00 [aio/6]
root 115 1 0 18:56 ? 00:00:00 [aio/7]
root 116 1 0 18:56 ? 00:00:00 [aio/8]
root 117 1 0 18:56 ? 00:00:00 [aio/9]
root 118 1 0 18:56 ? 00:00:00 [aio/10]
root 119 1 0 18:56 ? 00:00:00 [aio/11]
root 120 1 0 18:56 ? 00:00:00 [aio/12]
root 121 1 0 18:56 ? 00:00:00 [aio/13]
root 122 1 0 18:56 ? 00:00:00 [aio/14]
root 123 1 0 18:56 ? 00:00:00 [aio/15]
root 124 1 0 18:56 ? 00:00:00 [aio/16]
root 125 1 0 18:56 ? 00:00:00 [aio/17]
root 126 1 0 18:56 ? 00:00:00 [aio/18]
root 127 1 0 18:56 ? 00:00:00 [aio/19]
root 128 1 0 18:56 ? 00:00:00 [aio/20]
root 129 1 0 18:56 ? 00:00:00 [aio/21]
root 130 1 0 18:56 ? 00:00:00 [aio/22]
root 131 1 0 18:56 ? 00:00:00 [aio/23]
root 132 1 0 18:56 ? 00:00:00 [aio/24]
root 133 1 0 18:56 ? 00:00:00 [aio/25]
root 134 1 0 18:56 ? 00:00:00 [aio/26]
root 135 1 0 18:56 ? 00:00:00 [aio/27]
root 136 1 0 18:56 ? 00:00:00 [aio/28]
root 137 1 0 18:56 ? 00:00:00 [aio/29]
root 138 1 0 18:56 ? 00:00:00 [aio/30]
root 139 1 0 18:56 ? 00:00:00 [aio/31]
root 140 1 0 18:56 ? 00:00:00 [scsi_eh_0]
root 320 1 0 18:56 ? 00:00:00 /sbin/syslogd
root 323 1 0 18:56 ? 00:00:00 /sbin/klogd
root 330 1 0 18:56 ? 00:00:00 /usr/sbin/inetd
root 334 1 0 18:56 ? 00:00:00 /usr/sbin/netserver
root 340 1 0 18:56 ? 00:00:00 /usr/sbin/sshd
root 343 1 0 18:56 ? 00:00:00 /usr/sbin/ntpd
root 346 1 0 18:56 tty1 00:00:00 /sbin/getty 38400 tty1
root 347 1 0 18:56 tty2 00:00:00 /sbin/getty 38400 tty2
root 348 1 0 18:56 tty3 00:00:00 /sbin/getty 38400 tty3
root 349 1 0 18:56 tty4 00:00:00 /sbin/getty 38400 tty4
root 350 1 0 18:56 tty5 00:00:00 /sbin/getty 38400 tty5
root 351 1 0 18:56 tty6 00:00:00 /sbin/getty 38400 tty6
root 352 1 0 18:56 ttyS0 00:00:00 -bash
root 354 340 0 18:57 ? 00:00:00 /usr/sbin/sshd
wli 356 354 0 18:57 ? 00:00:00 /usr/sbin/sshd
wli 357 356 0 18:57 pts/0 00:00:00 -zsh
root 360 340 0 18:57 ? 00:00:00 /usr/sbin/sshd
wli 362 360 0 18:57 ? 00:00:00 /usr/sbin/sshd
wli 363 362 0 18:57 pts/1 00:00:00 -zsh
root 366 340 0 18:57 ? 00:00:00 /usr/sbin/sshd
wli 368 366 0 18:57 ? 00:00:00 /usr/sbin/sshd
wli 369 368 0 18:57 pts/2 00:00:00 -zsh
wli 4435 363 0 18:59 pts/1 00:00:00 ps -fade
wli 4436 363 0 18:59 pts/1 00:00:00 less
Hmm, is there any hope left for state machines?
Bill
P.S.: I'd also love to see make -j64 bzImage take 26s like on 16x
instead of 48s as it appears to be doing on 32x. Doubling
num_cpus_online() shouldn't double kernel kernel compile time.
Bill - so what happens if you trim down the aio, event and ksoftirqd
threads to a sane size (you might also want to do something about the
fact 2.5 still runs ksoftirq too easily). Intuitively I'd go for a
square root of the number of processors + 1 sort of function but what do
the benchmarks say ?
On Sun, Nov 17, 2002 at 12:11:35AM +0000, Alan Cox wrote:
> Bill - so what happens if you trim down the aio, event and ksoftirqd
> threads to a sane size (you might also want to do something about
> the fact 2.5 still runs ksoftirq too easily). Intuitively I'd go for
> a square root of the number of processors + 1 sort of function but
> what do the benchmarks say ?
IMO having various threads per-CPU is getting silly for (say) 4+
CPUs. Even for two CPUs it means quite a good number of kernel
threads.
Does anyone really know for certain that this is necessary versus
having few per-CPU threads calling into state-machine functions?
--cw
On Sun, Nov 17, 2002 at 12:11:35AM +0000, Alan Cox wrote:
> Bill - so what happens if you trim down the aio, event and ksoftirqd
> threads to a sane size (you might also want to do something about the
> fact 2.5 still runs ksoftirq too easily). Intuitively I'd go for a
> square root of the number of processors + 1 sort of function but what do
> the benchmarks say ?
Both reorganizing the per-cpu thread pools as state machines and
inserting new locking look like work-intensive projects...
It's not become explosively bad yet (1MB of overhead is eyebrow-raising
but not particularly damaging) so there's no rush to trim this down,
but I'm at least thinking about doing this later. One of the major
obstacles for the state machine approach is that the migration threads
run at RT priority while the rest do not, and of course the greater
than per-cpu granularity approach suffers from additional locking.
Bill